Link prediction in social networks using network analysis methods and long short-term memory

Link Prediction (LP) is a prominent task in social and complex network analysis aiming to predict new connections or edges in the network based on its current features and structure. LP plays a crucial role in diverse applications, showcasing its significance in enhancing connectivity and estimating the growth and evolution of networks. Existing methods of LP commonly rely on local, semi-local, or global network features, often encountering issues of inconsistent performance across diverse and large-scale networks. This paper proposes an innovative and refined approach to link prediction in complex and static networks by integrating network analysis methods, namely, similarity methods and node centralities, and a Long Short-Term Memory (LSTM) classifier. The proposed model computes PageRank and closeness centrality measures for individual nodes. This is followed by computing similarity measures such as Leicht–Holme–Newman (LHN) Similarity for every node pair in the network. The fusion of popularity and similarity measures generates some combined features for each node pair, these combined features, accompanied by positive or negative edge labels, establish a well-defined dataset for link prediction. Subsequently, the labeled dataset is employed by LSTM classifiers toward the link prediction task. An extensive experiment and simulation are conducted on seven diverse social and complex network datasets to evaluate the performance of the proposed model in comparison to several existing link prediction methods. The results demonstrate the improved performance of the proposed approach.